Internet Engineering Task Force Jiri Kuthan
Internet Draft GMD Fokus
draft-kuthan-midcom-framework-00.txt Jonathan Rosenberg
November, 2000 dynamicsoft
Expires: May 2001
Middlebox Communication: Framework and Requirements
Status of this Memo
This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026 [1].
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETFi), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts.
Internet-Drafts are draft documents valid for a maximum of six
months and may be updated, replaced, or obsoleted by other documents
at any time. It is inappropriate to use Internet- Drafts as
reference material or to cite them other than as "work in progress."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html.
Abstract
The purpose of this document is to develop framework and
requirements for a protocol that will allow for communicating
control data associated with IP/transport-layer data flows or
aggregates of them between intermediate packet processing devices
and external controllers.
The protocol will be extensible in order to allow for communicating
arbitrary control data associated with packet flows and defining
packet flow processing. It will include provisions for verifying the
integrity of each message as well as ensuring authentication of all
parties involved in the transactionsi.
A major application of this protocol will be the control of packet
processing policies in decomposed firewalls/NATs/NATi-PTs by
externalized Application Level Gateways. This particular use will
relieve firewalls/NATs from application-layer processing to improve
their maintainability and performance.
Examples of other possible applications include but are not limited
to buffer management and load balancing.
J. Kuthan, J. Rosenberg [Page 1]
^LInternet Draft Middlebox Framework and Requirements November 2000
Contents
1 Introduction.....................................................2
2 Terminology......................................................3
3 Case Study: Traversal of Applications Using Session Control
Protocols across Firewalls/NATs ..................................5
4 Protocol Requirements for FCP....................................7
4.1 FCP Operational Model..........................................7
4.2 Functional Requirements: Management of Packet Flow Processing
States .........................................................7
4.3 Rule Manipulation Operations...................................7
4.4 Soft-state Rule Design.........................................7
4.5 Rule Language..................................................7
4.5.1 Packet Matching Expressions..................................8
4.5.2 Rule Processing Precedence...................................8
4.5.3 Control State Content........................................9
4.6 Feedback......................................................10
4.7 Security......................................................11
4.8 Reliability...................................................11
4.9 Real-time Operation...........................................11
4.10 Extensibility................................................11
4.11 On Support Specific to NAT/NAPTi/NAT-PT.......................12
5 Related Issues..................................................13
5.1 Access Control................................................13
5.2 Rule Ownership................................................13
5.3 Default Flows.................................................13
5.4 Location of FCP Controllers...................................14
6 Open Issues.....................................................14
6.1 Location of Intermediate Devices..............................14
7 Performance and Scalability Considerations......................15
8 Security Considerations.........................................15
9 Document Status.................................................15
10 Acknowledgments................................................16
1 Introduction
Though the Internet has been designed to provide network layer
connectivity end to end [2] it actually consists of mixed network
realms. These include IPv4 networks, IPv6 networks, networks hidden
behind NATs and firewalls. This problem was referred to as
"transparency loss" and discussed in [3]. Applications being run
across mixed realms may experience lack of interoperability or
suboptimal performance.
To assists applications in traversing network boundaries Application
Level Gateways (ALGs) embedded in intermediate devices have been
used. However, tight coupling of application and network/transport
layer processing results in reduced maintainability of the
intermediate devices. Built-in application awareness typically
requires updates of operating systems with new applications or
versions of it. Operators of such systems wanting to support new
J. Kuthan, J. Rosenberg [Page 2]
^LInternet Draft Middlebox Framework and Requirements November 2000
applications cannot use software supplied by third parties and are
at the mercy of vendors of their equipment. Furthermore, support of
numerous application protocols increases complexity of such
integrated systems and may affect performance.
To deal with this sort of problems we suggest decomposition of
application awareness from network/transport layer processing. We
assume that applications control network/transport layer processing
in intermediate devices through a generic application-independent
control protocol. In the common case, the application-awareness
resides in proxies ("externalized ALGs") that hide boundary
traversals from end-devices.
+--------------------+
+------+-----------+ |
+----------+ FCP | | Per-Flow | |
+----------+|..........| | State | |
| ALGs ||..........| FCP | Table | | policy +--------+
+----------+ | unit |-----------| | protocol | policy |
| | ACL ~~~~~~~~~~~~~~+ server |
__________+------+-----------+ | +--------+
| Intermediate |
| Device |
+--------------------+
Device | |
Interfaces | | ...
IN OUT ...
Legend: .... FCP
~~~~ policy protocol
Figure 1: Middlebox Communication Architecture
The rest of this document describes framework and requirements for
the missing piece, the control protocol. We refer to this protocol
as Flow Processing Control Protocol (FCP). Discussion on how FCP
maps or does not map to an existing protocol is out of scope of this
document. Section 2 defines terms used throughout this memo. In
Section 3, we demonstrate the use of the FCP in a case study. We
formulate protocol requirements in Section 4. Issues that are
related to protocol operation but do not affect protocol
specification are summarized in Section 5. Unresolved issues are
identified in Section 6. Section 7 reviews performance issues.
Security is considered in Section 8 and status of this memo in
Section 9.
2 Terminology
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALLNOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in
this document are to be interpreted as described in RFC 2119.
J. Kuthan, J. Rosenberg [Page 3]
^LInternet Draft Middlebox Framework and Requirements November 2000
o Endpoint address - general term describing source or destination
of a packet. This is, depending on context, IP address and/or
TCP/UDP port number.
o Packet matching expression - expression that specifies values of
header fields of packets to be selected. The inspected header
fields MAY be but are not limited to source and destination IP
addresses, TCP/UDP port numbers, etc.
o Packet flow - a sequence of packets matched by a packet matching
expression.
o Rule - a packet matching expression and control state used to
determine processing of packets matched by the expression.
o Session - a set of packet flows belonging to an application.
o Session control protocol - a protocol used to negotiate endpoint
addresses of flows belonging to a session. Examples of such
protocols are SIP, H.323, RTSPi.
o Proxy - an intermediary server that relays application messages
from one entity to another one. A proxy acts as server for message
senders and as client on behalf of the senders for message
receivers. It may be used for enforcement of application-level
policies such as content filtering or message translation. With
applications using session control protocols, proxies typically
relay session control protocols and do not relay data flows
belonging to a session.
o Packet filter - a network device that examines headers of
forwarded packets and allows only packets conforming to a security
policy to pass through. The security policy defines which endpoint
addresses are considered trustworthy and which are not. For
example, it may permit data traffic of an application identified
by a port numbe only from/to a trusted proxy.
o Network Address Translator (NAT) - a packet processing device that
is able to map source and/or destination endpoint addresses of
forwarded packet flows to a pool of other addresses. This
technique is used to conserve IP address space and/or hide IP
address of hosts behind the NATs from outside of the NATs. The NAT
concept is described in [10].
o Bind - a pair consisting of an original and translated endpoint
address.
o Intermediate device - a packet-processing device located along
end-to-end path. It may be a packet filter, NAT, intrusion
detection system, load balancer, etc.
o Firewall - centrally maintained devices set-up to increase network
security by putting restrictions on information flows. The
restrictions are applied with packet filters at the packet level
and/or proxies at the application level. Optionally, NATs may be
used. Note that the term "firewall" is sometimes used to refer to
packet filters.
o Application Level Gateways (ALGs) - application-aware modules that
control processing state in firewalls/NATs and manipulate
application messages to accomplish firewall/NAT traversal.
Typically, ALGs are embedded in firewalls/NATs. They may also be
J. Kuthan, J. Rosenberg [Page 4]
^LInternet Draft Middlebox Framework and Requirements November 2000
externalized to remote proxies if a control interface between them
and firewalls/NATs is provided [4].
o Flow Processing Control Protocol (FCP) - protocol for
communicating flow processing policy between external controllers
and intermediate devices. The intermediate devices accept only
authorized FCP requests. The authorization can come from an
internal access control list or an external policy server.
o Access Control List (ACL) - policy defining who may
access/manipulate controlled devices with FCP. The ACLs may be
outsourced to an external policy server.
3 Case Study: Traversal of Applications Using Session Control Protocols
across Firewalls/NATs
Firewalls are trusted, administrator-maintained devices used to
increase network security by enforcing restrictions on information
flows. The restriction policies are centrally defined and maintained
by network administrators. The firewalls consist of proxies and
packet filters. Proxies are application-aware entities acting on
behalf of untrusted hosts at application layer. They examine
application protocol flows and allow only messages conformant to
security policies to pass through. Optionally, they modify the
messages to make them policy-conformant. Packet filters are used to
impose security restrictions at lower layers. They usually inspect
IP and TCP/UDP packet headers against tables of filtering rules.
Only conforming IP packets are allowed to pass through filters. The
packet filtering policy may be either 'default-permit' or 'default-
deny'. 'Default-permit' policy allows all but explicitly stated IP
flows whereas 'default-deny' policy allows only explicitly stated IP
flows to pass through. Typically, the latter policy is set up to
allow traffic from and to trusted proxies to pass through. It
provides higher security by being more restrictive. Thus, it is
frequently deployed in corporate networks.
Unfortunately, this policy makes firewall traversal difficult for
applications using session bundles. This means that such
applications (e.g., SIP [5], H.323 [6], RTSP[7], and FTP
[8])negotiate IP addresses and port numbers with a session control
protocol dynamically and then use the negotiated addresses to
establish packet streams for transport of data. This technique is
useful, for example, if multiple applications want to receive RTPi
[9] flows and cannot share the same TCP/UDP port number or an
application uses port numbers to demultiplex multiple incoming RTP
flows. It is also necessary if IP address information is dynamic.
As a result, dynamically created sessions fail to traverse firewalls
deploying static 'default-deny' filtering policies. If network
address translation (NAT)[10] is deployed the traversal fails as
well because session signaling conveys unroutable IP addresses and
port numbers. This problem has been addressed in firewalls/NATs by
usage of embedded Application Level Gateways (ALGs) [15]. They adapt
J. Kuthan, J. Rosenberg [Page 5]
^LInternet Draft Middlebox Framework and Requirements November 2000
packet processing policy to application needs dynamically. However,
embedded ALGs suffer from limited maintainability, increase
complexity of firewalls/NATs and fail to operate if message
authentication or encryption is used.
Thus we suggest using external application proxies that control
firewalls/NATs and relieve them from application processing.
Arbitrary application proxies may be added to manageable firewalls
easily. Hop-by-hop security is enabled. Existing software such as
SIP proxies or H.323 gatekeepers may be used without duplicating the
applications' protocol stacks in the firewalls/NATs. The reference
scheme is depicted in Figure 2:. There are FCP-enabled, trusted,
administrator-maintained proxies acting as external ALGs and
controlling a packet filter located within the same administrative
domain. The packet filter implements the 'default-deny' packet
filtering policy. It permits session control traffic from and to the
trusted proxies and accepts FCP requests from them. The proxies use
their application-awareness to control the packet filter dynamically
with FCP. They also enforce application-level policy such as
dropping messages infected with known viruses, or lacking user
authentication. The policy decisions may be delegated to an external
policy server.
+--------+
| App. |
| Policy | +---------+ SIP
| Server |~~~~~~~~~| SIP +_____________ |
+--------+ ________| Proxy | \ |
/ +---------+.. +----+---------------+
| : FCP +------+-----------+ |__
| RSTP +----------+ :...........| | Per-Flow | |__
SIP | ____| RSTP |..............| | State | |__
| / | Proxy |______________| FCP | Table | |
| | +----------+ | unit | -------- | |
| | FTP +---------+.............| | ACL | |
| | _____|FTP Proxy|_____________+------+-----------+ |
| | / +---------+ | Packet |
| | | -----| Filter |--
+-----------+ /-----| |--
+-----------+| data streams // +----+---------------+
+-----------+||----------->----// |
|end-devices||------------<----- |
+-----------+ (RTP, ftp-data, etc.) |
Inside | Outside
Legend: ---- raw data streams
____ application control protocols
.... FCP
~~~~ policy protocol
Figure 2: Use of FCP for Firewall/NAT Decomposition
J. Kuthan, J. Rosenberg [Page 6]
^LInternet Draft Middlebox Framework and Requirements November 2000
4 Protocol Requirements for FCP
4.1 FCP Operational Model
In the remainder of this document we assume a model in which packet
processing in an intermediate device located at a network edge is
determined by an ordered set of rules. The rules consist of packet
matching expressions and control state. The packet matching
expressions select packets and associated control state determines
how selected packets are treated. A packet may match multiple packet
matching expressions. If this happens the first one will be taken.
The rules are manipulated with FCP dynamically. Multiple FCP
controllers MAY control a single intermediate device. It is expected
that only a few trusted hosts from a single administrative domain
will act as FCP controllers.
4.2 Functional Requirements: Management of Packet Flow Processing
States
The primary goal of FCP is to allow for remote dynamic management of
packet flow processing rules. As a minimum requirement, the FCP MUST
enable controllers to permit/forbid processing of specific packet
flows and request/release NAT/NAPT/NAT-PT[11] translations.
4.3 Rule Manipulation Operations
FCP MUST allow for setting, releasing and querying packet flow
processing rules. Operations like modification of existing rules and
keeping them alive are most likely to be implemented with the 'set'
operation.
The 'set' operation MAY either specify values of updated state
elements explicitly or omit them to allow controlled entities to
assign appropriate values. These MAY be default values (e.g. 0 for
packet counter), ephemeral values, or current values if the state
elements already exists (useful for keep-alive messages).
4.4 Soft-state Rule Design.
To avoid accumulation of stale rules in case of controller failures
rules MUST have timers that expire if they are no more refreshed by
controllers. FCP MUST enable controllers to refresh rules
periodically. FCP MUST also allow controllers to set the timer's
length -- it is frequently a controller that knows best what timer
length is appropriate. If a controller does not specify timer value
explicitly a default value will be assigned. A trivial value for
infinite timer MUST be defined.
4.5 Rule Language
J. Kuthan, J. Rosenberg [Page 7]
^LInternet Draft Middlebox Framework and Requirements November 2000
This section specifies requirements for the language of packet
rules. Note that FCP-controlled hosts have to understand all
expressions written in this language but FCP controllers may use
only a subset of them.
4.5.1 Packet Matching Expressions
A) As a minimum requirement, the language of packet matching
expressions MUST allow for specification of the following
protocols and their respective header fields:
- IPv4: source and destination IP address or group of them
determined by a netmask, protocol number, TOS field
- IPv6: source and destination IP address or group of them
determined by a netmask, next header fields (Note that multiple
fields may need to be traversed until a match is found.),
traffic class field
- UDP: source and destination port numbers or group of them
- TCP: source and destination port numbers or group of them, "SYN
packets" permission
- ICMPi: type and code
- IGMP: type
B) FCP controllers MUST be able to specify in which direction
packets may traverse controlled devices. This requires notion of
device interfaces. The notion of interface is abstract and
independent on interface technology and assigned IP address.
Support for generic predefined interface names "in", "out",
"loopback" (synonym for senders and receivers located at the
controlled device), and "DMZ" (demilitarized zone) is REQUIRED.
Packet traversal direction may be expressed in various ways, for
example by inbound and outbound interfaces or by interface and
direction in which packets pass it.
4.5.2 Rule Processing Precedence
The ability to indicate desired rule processing precedence is
REQUIRED to enable controlled devices to resolve conflicts between
multiple applicable matching rules in a predictable manner. If no
precedence is specified for a rule, default precedence will be
assigned by FCP-controlled device. Multiple rules MAY share a single
precedence.
Note: Though precedence sharing leaves processing order of rules
with the same precedence undetermined it greatly simplifies
certain common cases. For example, allocating a single precedence
for all dynamically generated firewall pinholes does not affect
firewall's behavior because all the pinhole rules result in the
same action, which is packet forwarding. Then none of multiple FCP
controllers needs to determine at which position a new rule will
be inserted in a rule base.
J. Kuthan, J. Rosenberg [Page 8]
^LInternet Draft Middlebox Framework and Requirements November 2000
4.5.3 Control State Content
The control state associated with a packet matching expression in a
rule keeps information related to a packet flow. It MAY include flow
processing actions, timer information, number of matched packets,
traffic limitations, etc. Members of the control states are subject
to future extensions.
The following control state members are REQUIRED:
A) "Action" defines whether matched packets are forwarded. It takes
the values 'pass packets', 'drop packets with or without ICMP
notification'.
B) "Rule timer" defines time remaining until state expiration. See
also discussion of soft-state rule design.
C) "Logging" of asynchronous events related to a rule. The protocol
MUST allow FCP controllers to request logging of asynchronous
events such as packet match and timer expiration. The protocol
MUST enable controllers to specify log level and frequency. The
log frequency is used to avoid voluminous logging if an event
occurs frequently. Choice of the notification/logging mechanism
is a configuration option that does not need to be specified with
FCP.
D) Unique "flow state identifiers" are REQUIRED unless matching
expressions are uniquely identifiable. Otherwise, state
modification/releasing could not work consistently. The
identifiers may be generated either by controllers or by
controlled devices. They may be random or ephemeral. If
controllers generate them, they MUST be random to avoid
collisions with identifiers generated by other controllers. If
controlled devices generate identifiers, they MAY be ephemeral.
Ephemeral identifiers are typically shorter but lose their
uniqueness under a failure.
E) "Packet modifier" allows to describe one or more rules for re-
writing header fields of matched accepted packets. The modifier
rules will consist of identification of the packet fields to be
changed, operators (at least the assignment operator is REQUIRED)
and operands. In particular, the modifier MUST be able to change
the following protocol header fields:
- IPv4: IP addresses, TOS field
- IPv6: IP addresses, traffic class field
- UDP: port numbers
- TCP: port numbers
Note that if modifiers are used packet checksums MUST be
recalculated.
J. Kuthan, J. Rosenberg [Page 9]
^LInternet Draft Middlebox Framework and Requirements November 2000
The following control state members are RECOMMENDED:
F) "Packet counter" keeps number of packets matched by a rule.
G) "Maximum packets per second" specifies the maximum allowed packet
rate of a flow. Packets exceeding this rate will be dropped.
H) "Maximum bitrate" specifies the maximum allowed bitrate of a
flow. Packets exceeding this rate will be dropped.
I) "Inactivity timer" specifies period of time after which a rule is
released if no packet matches.
J) "Reflexive Rules": In order to allow replies to TCP/UDP data
flows originated from the internal side of firewall while still
keeping the filtering policy as restrictive as possible, so-
called reflexive rules are used. Reflexive rules are rules that
allow packet flows reverse to explicitly permitted active flows.
They are defined implicitly by permitting their generation within
specification of the original explicit rules. They specify the
same protocol, IP addresses, port numbers as flows matching the
original rule do except the addresses and port numbers are
swapped. If permitted, packet filters generate a reflexive rule
whenever a new flow matches an explicit rule. No controller's
intervention is needed. The reflexive rules are valid only
temporarily. They MUST be released when an inactivity timer
expires, the flow is terminated explicitly (e.g., by a FIN
segment with TCP) or the original rule is deleted. If creation of
reflexive rules is supported, FCP MUST allow to specify values of
their control state members.
Note: If endpoint address modifiers are enabled in the original
rules they MUST be reflected in the reflexive rules; namely
packet-matching expressions of the reflexive rules MUST match
flows reverse to modified flows and modifiers MUST be enabled to
translate endpoint addresses of reverse flow to addresses before
modification.
4.6 Feedback
FCP controllers need to receive feedback on their control messages
in order to learn about results of requested operations. Both
positive and negative responses are REQUIRED.
Positive responses indicate successful operation and MAY possibly
describe content of the controlled states or part of them. Per-flow
control state or part of it is always returned if it was asked for
by 'query' operation.
J. Kuthan, J. Rosenberg [Page 10]
^LInternet Draft Middlebox Framework and Requirements November 2000
Negative responses indicate failures and describe reasons for them.
Minimum negative responses REQUIRED are "Authentication failed",
"Not permitted", "Syntax Error".
4.7 Security
In order to prevent unauthorized entities from manipulating the
state of controlled device the FCP channel MUST be secure. It MUST
include provisions for verifying the integrity of each message as
well as ensuring authentication of all parties involved in the
transactioni.
It is RECOMMENDED that the FCP channel is private so that a
malicious listener cannot find out packet processing policy easily.
The security protocols may take place either at lower layers (IPSec,
TSL) or at the FCP layer.
Though IP-address based authentication may be satisfactory in
particular cases cryptographic authentication is REQUIRED generally.
Note that we discuss the security between FCP controllers and
controlled device. Security mechanisms used by applications to
communicate with FCP controllers are a separate issue out of
scope of this memo.
4.8 Reliability
As with almost any other control protocol reliability is REQUIRED
regardless if it is implemented by FCP itself or an underlying
transport protocol.
4.9 Real-time Operation
The protocol transactions must be fast in terms of RTT to avoid
introducing delays to applications. Unless network loss results in
retransmissioni, total transaction time SHOULD be one RTT.
Note: If TCP is used as underlying protocol to provide
reliability, pre-established TCP connections may be used to
reduce transaction time.
4.10 Extensibility
Protocol extensibility is REQUIRED in order to enable reuse of FCP
for control of a variety of packet-processing mechanisms. In
particular, adding new control state fields (e.g., buffer management
information), new reply codes and elements of packet matching
expression language MUST be supported.
J. Kuthan, J. Rosenberg [Page 11]
^LInternet Draft Middlebox Framework and Requirements November 2000
The protocol MUST convey protocol version number in order to make
transition to potential future versions easier.
4.11 On Support Specific to NAT/NAPT/NAT-PT
One of the FCP purposes is to communicate NAT/NAPT/NAT-PT binds
between controllers and controlled devices. Knowledge of the binds
is necessarily needed by session control proxies to operate
properly.
The primary question is who creates the binds. One alternative is
controllers request a new bind and NATs create and return it. The
other choice is the controllers create a bind and instruct NATs to
use it.
Locating the bind logic in NATs follows the decomposition concept
"IP/transport intelligence in controlled devices, application
intelligence out of them". It relieves controllers such as SIP
proxies from maintenance of the address pools and making bind
assignments. It avoids collisions that would be due if multiple
controllers would access a single device. (We do not consider
splitting a pool of public addresses among multiple controllers a
solution. It would beat the purpose of NAT which is address
sharing.)
A minor drawback of this logic placement is it requires two-stage
transactions if NATs are co-located with firewalls. In the first
stage, a controller must find out, if NAT applies to a given address
and request a bind to include in its signaling. In the second stage,
when application signaling is over, it permits a packet flow using
the reserved translation. With bind logic residing in controllers,
both operations may be done jointly in the second phase and the
first phase can be skipped.
A specific scenario where locating the bind logic to controllers is
advantageous is if a controller wants to make sure the same address
translation is applied in multiple controlled devices. Clearly, this
would not be possible if the devices assigned the binds
independently.
We leave the answer to location of bind intelligence to
configuration. It is REQUIRED that FCP supports both alternatives.
The following protocol operations are REQUIRED:
o FCP controllers MUST be able to request NAT translations. If NAT
is used, controlled devices allocate an address translation and
return it.
o FCP controllers MUST be able to set NAT translations. (Note that
this can be accomplished with modifiers.) Controlled devices MUST
be able to indicate if the translation cannot be set because it is
already reserved.
J. Kuthan, J. Rosenberg [Page 12]
^LInternet Draft Middlebox Framework and Requirements November 2000
o With NAPT, allocating port blocks is REQUIRED, i.e., FCP
controllers MUST be able to request a translation of a contiguous
block of port numbers and controlled devices MUST allocate a
contiguous block of translated port numbers to such request. The
least significant bit of both private and translated port numbers
MUST be same. It is needed, for example, by RTP [9], which
recommends allocation of even port numbers for itself, subsequent
port number for RTCPi and contiguous block of port numbers for
layered encoding applications.
o The controllers MUST be able to release the allocated bindings.
o The allocated address bindings are subject to timer expiration in
a similar way as soft-state packet-processing rules are.
5 Related Issues
This Section explicitly names related features that are out of scope
of protocol requirements and are matter of implementation,
administrative policy or anything else.
5.1 Access Control
There may be different access control lists (ACLs) defining who may
access and modify what rules in an intermediate device. For example,
an ACL may specify that an FCP controller may only control rules
describing traffic to and from a specific subnet. Additionally, it
may define in which way the controller is required to authenticate
and which precedence it may use for its rules. The access control
policies may be stored and applied locally or they may be outsourced
to an external policy server using a policy protocol. In either case
they are out of scope of FCP. The only required FCP feature is
authentication support.
5.2 Rule Ownership
Multiple controllers may control a single device with FCP. It is
desirable to avoid modification of per-flow control states by other
entities than those that created them (perhaps with exception of a
network manager). Thus, the controlled devices MUST implement the
notion of rule ownership. The only required protocol functionality
is authentication.
5.3 Default Flows
If a packet does not match any of matching expressions maintained in
a packet filter a default rule has to be applied. Otherwise, packet
handling would be undefined. Thus, all packet filters controlled by
FCP must always maintain the default rule. The matching expression
of the rule matches all packets at lowest priority so that any other
matching rules take precedence over it. The content of the default
control state MAY be modified with FCP, the matching expression MUST
NOT.
J. Kuthan, J. Rosenberg [Page 13]
^LInternet Draft Middlebox Framework and Requirements November 2000
5.4 Location of FCP Controllers
FCP controllers may be located on any side of controlled network
device. Their location with respect to the controlled device does
not affect protocol specification but may result in different
protocol flows. For example, an application proxy located on the
private side of a NAT needs to set up a single permanent translation
that enables it to receive inbound messages and forward them to
their destinations. If the proxy is located on the public side, it
needs to set up multiple translations for inbound messages forwarded
to individual destinations located on the private side.
6 Open Issues
6.1 Location of Intermediate Devices
Determining which intermediate device a controller should control is
out of the scope of this document. Administrators can accomplish
this task manually. Alternatively, a discovery protocol could be
used.
A difficult problem arises if packet flows may take path through
multiple intermediate devices at the network edge. FCP controllers
cannot easily determine which of them they should control. The
problem is illustrated in the example depicted in Figure 3:
IN | OUT
+-----+ SIP +------------+ +-------+
| SIP |____________| firewall 1 |____________| SIP |
|proxy|............| | | proxy |
+-----+ : +------------+ +-------+
| : FCP ... | |
| MGCPi : +------------+ MGCP |
| :.......|firewall i | |
+--------+ : +------------+ +-------+
|media | : ... | ?<-------|media |
|gateway |--->? : +------------+ |gateway|
+--------+ :..|firewall N | +-------+
+------------+
|
Figure 3: Controlling Multiple Intermediate Devices
In this example, multiple firewalls 1 .. N are present in a network.
A SIP proxy relays SIP signaling, has knowledge of all the firewalls
and is authorized to control them. It knows source and destination
endpoints of data flows belonging to a session but does not know
which of the firewalls they will traverse. It cannot calculate it
J. Kuthan, J. Rosenberg [Page 14]
^LInternet Draft Middlebox Framework and Requirements November 2000
because it does not know routing tables along the entire end-to-end
path.
Solutions are still sought. A possible solution is to let
controllers instruct all controlled devices in parallel, most likely
using multicast. This solution scales only for a small number of
controlled devices. With NAT, it assumes the translation assignments
to be communicated from FCP controllers to controlled devices.
7 Performance and Scalability Considerations
The ability to add processing rules to control packet-processing
devices dynamically may result in creation of large and rapidly
changing rule tables. For example, if FCP is used to open pinholes
in a 'default-deny-and-dynamic-open' firewall for Internet telephony
sessions the table size grows with number of sessions linearly. The
lookup processing overhead grows as well and may lead to increased
packet latency. Maintenance of per-flow states makes use of FCP
meaningful only in network edges.
Mechanisms for fast rule lookup in large, frequently changing filter
databases are needed. Results of some recent research in this area
were published in [12], [13], and [14]. Use of packet modification
may also affect processing performance.
A performance improvement may be reached administratively by
definition of an application-aware rule precedence policy. A
controller may request that rules for packet flows with higher
expected packet rate will be assigned a higher precedence than rules
for packet flows with lower packet rate. Then, the most commonly
accessed rules will be processed first and average packet processing
time will decrease. Note that this mechanism is not extremely fair
to streams with low bandwidth consumption since their processing
time will increase.
8 Security Considerations
The security requirements for the control protocol are described in
Section 4.7. Note that security of the protocol does not help alone.
Additionally, security of the entire control system is subject to
security of the FCP controllers and access control in FCP-controlled
devices (see Section 5.1.).
9 Document Status
This document is in the stage of collection of requirements and open
issues. Numerous updates result from discussions on the foglamps
mailing list. Previous versions were issued as draft-kuthan-fcp-
{00|01}.txt.
J. Kuthan, J. Rosenberg [Page 15]
^LInternet Draft Middlebox Framework and Requirements November 2000
10 Acknowledgments
Numerous people have been contributing to collection of these
requirements. Many document clarifications and enhancements resulted
from discussions on the foglamps mailing list. We specially
acknowledge the following people for their help: Scott Bradner,
Stefan Foeckel, Melinda Shore, Dave Oran, and Jon Peterson. The
firewall traversal problem was stated in [15], [16].
Appendixes
A Examples
This section shows how to use FCP by examples. Many of the examples
refer to the application described in Section 3 and use SIP as a
prominent example of a session control protocol.
A.1 FCP Transaction Examples
This section illustrates how FCP requests could look like. The
requests in the following examples use abstract syntax in this form:
PME=
[ [=] ...]
The syntax of packet matching expression is borrowed from tcpdump.
An additional keyword 'if' specifies interface to whose incoming
queue the matching expression is applied. A similar syntax is used
for definition of packet modifiers. Discussion on how these abstract
FCP examples map or do not map to existing protocols is out of scope
of this document.
In the examples bellow, a protected host behind a firewall has the
address 10.1.1.1, an outside host has the address 130.149.17.15 and
the firewall's outbound interface has 193.174.152.25.
Example 1: Opening a Pinhole in a Packet Filter for an Outgoing Flow
In this example, a controller opens a pinhole for a packet flow
being sent from the inside to the outside host.
SET
PME='if in and udp src port 55 and src host 10.1.1.1 and udp dst
port 77 and dst host 130.149.17.15'
action=pass
=> REPLY: OK
Example 2: Opening a Pinhole in a Packet Filter w/NAPT for an
Outgoing Flow
J. Kuthan, J. Rosenberg [Page 16]
^LInternet Draft Middlebox Framework and Requirements November 2000
In this example, a controller queries a NAT bind and opens a pinhole
for a translated packet flow being sent from the outside to the
inside host through a NAT.
QUERY_NAT_TRANSLATION
udp:10.1.1.1:55
=> REPLY: NAT_OK, udp:10.1.1.1:55=udp:193.174.152.25:48374
SET
PME='dst host 10.1.1.1 and udp dst port 55 and if out and src host
130.149.17.15 and udp src 77'
action=PASS
=> REPLY: OK
Example 3: TOS Control
The controller instructs the controlled device to set TOS of matched
packets to hexadecimal value 0x10.
SET
PME='if in and udp src port 55 and src host 10.1.1.1 and udp dst
port 77 and dst host 130.149.17.15'
modifier='tos=0x10'
=> REPLY: OK
Example 4: Querying Number of Matched Packets
QUERY
PME='if in and udp src port 55 and src host 10.1.1.1 and udp dst
port 77 and dst host 130.149.17.15'
packet_count
=> REPLY: OK, packet_count=333
Example 5: Refreshing Per-Flow State
SET
PME='if in and udp src port 55 and src host 10.1.1.1 and udp dst
port 77 and dst host 130.149.17.15'
=> REPLY: OK
Example 6: Network Ingress Filtering
See [17] for more details on this scenario. The first rule denies
all packets on the "in" interface. The second rule with higher
priority explicitly permits packets from the 10.1.2 network.
J. Kuthan, J. Rosenberg [Page 17]
^LInternet Draft Middlebox Framework and Requirements November 2000
SET
PME='if in'
precedence=default
action=drop(no_ICMP)
=> REPLY: OK
SET
PME='if in and src net 10.1.2'
precedence=high
action=pass
=> REPLY: OK
Example 7: Reflexive HTTP Rules
The next rule allows controlled packet filters to create temporary
rules that permit inbound TCP packets for HTTP transactions
initiated from the internal side of a firewall.
SET
PME='if in and tcp dst port 80'
REFLEXIVE='permit=yes, timer=180s, if=out'
=> REPLY: OK
If an HTTP request from 10.1.1.1:37313 to 130.149.17.15:80 matches
this rule a reflexive rule of the following form is generated:
PME='if=out and tcp src port 80 and src host 130.149.17.15 and tcp
dst port 37313 and dst host 10.1.1.1'
Example 8: Packet Redirection
In this scenario, all HTTP traffic from inside network is redirected
to a Web proxy (10.1.4.4) transparently. This scenario is sometimes
also referred as 'transparent proxy'. The rule allows for automatic
creation of reflexive rules.
SET
PME='if in and tcp dst port 80'
modifier='ip dst host = 10.1.4.4'
reflexive_rules='permit=yes, inactivity_timer=240s, if=dmz'
=> REPLY: OK
If an HTTP request from 10.1.1.1:37313 to 130.149.17.15:80 matches
this rule a reflexive rule of the following form is generated:
PME='if=dmz and tcp src port 80 and src host 10.1.4.4 and tcp dst
port 37313 and dst host 10.1.1.1'
J. Kuthan, J. Rosenberg [Page 18]
^LInternet Draft Middlebox Framework and Requirements November 2000
modifier='ip src host=130.149.17.15
A.2 Using FCP to Get an Outgoing SIP/SDPi Session through a 'Default-
Deny' Firewall w/NAPT
This example illustrates how FCP can be used to get an outgoing SIP
call through a firewall deploying 'default-deny' packet filtering
policy. Network configuration as displayed in Figure 1 is assumed:
The packet filter allows SIP signaling only from/to a SIP proxy, the
proxy rejects calls considered untrustworthy, and instructs the
packet filter to open pinholes for RTP streams belonging to
trustworthy SIP/SDP sessions for the time of session duration.
Additionally, NAPT is deployed.
Precise timing of opening and closing pinholes in SIP sessions and
issues such as 183 provisional media and re-invites are subject to
discussion which is out of scope of this document. Management of
RTCP and ICMP pinholes is omitted for the sake of simplification.
Note that the pinholes in the packet filter are quite 'wide'. This
means they allow packets with arbitrary source address and port
number to pass through because SDP does not communicate source
endpoint addresses.
Notation: In the diagram "INV 10.1.1.1:55" means an INVITE message
with the SDP body indicating IP address 10.1.1.1 with port 55 as the
receiving address and port for an incoming media-stream. Similarly
"200 OK 130.149.17.15:77" indicates an OK response with IP address
130.149.17.15 and port 77 for receiving media. The value 0.0.0.0:0
stands for any IP address and port number. Per-flow control states
in this example are identified by packet matching expressions.
+---------------------------------------------+--------------------+
| INSIDE | OUTSIDE |
+---------------------------------------------+--------------------+
10.1.1.1 193.174.152.25 130.149.17.15
UACi SIP Proxy AuthServer NAT/FW UASi
| | | | |
| | | | |
/* process SIP invitation, bind to a public address for receiving
media, modify the invitation accordingly; do not open firewall
pinholes until both parties agree to establish a call; note
that binding of source address for outgoing media is not done
because SDP does not care about source addresses */
| ----------------->| | | |
| INV 10.1.1.1:55 | | | |
| | ------> | | |
| | auth ? | | |
| | <------ | | |
J. Kuthan, J. Rosenberg [Page 19]
^LInternet Draft Middlebox Framework and Requirements November 2000
| | OK auth | | |
| | | | |
| | ----------------------> | |
| |FCP query_nat | |
| | udp :10.1.1.1:55 | |
| | <---------------------- | |
| |OK udp:193.174.152.25:66 | |
| | -------------------------------------------> |
| | INV 193.174.152.25:66 |
/* process SIP OK, open NAT-enabled pinholes for outgoing and
incoming media as soon as SIP ACK arrives */
| | <------------------------------------------- |
| | 200 OK 130.149.17.15:77 |
| <-----------------| | |
| 200 OK 130.149.17.15:77 | |
| ----------------->|--------------------------------------------->|
| ACK | ----------------------> | |
| |FCP SET | |
| |PME='dst udp 130.149.17.15:77 |
| | src udp 0.0.0.0:0 | |
| | if=in', action=PASS | |
| | <---------------------- | |
| | FCP OK | |
| | ----------------------> | |
| |FCP SET | |
| |PME='dst udp 10.1.1.1:55 | |
| | src udp 0.0.0.0:0 | |
| | if=out', action=PASS | |
| | <---------------------- | |
| | FCP OK | |
| | -------------------------------------------> |
| | ACK | |
| ...............................................................> |
| UDP/RTP DST 130.149.17.15:77 |
| <...........................................~................... |
| UDP/RTP DST 10.1.1.1:55 UDP/RTP DST 193.174.152.25:66|
/* close pinholes when either party sends SIP BYE */
| | <------------------------------------------- |
| <---------------- | BYE | |
| BYE | | |
| ----------------->| | |
| 200 OK | ----------------------> | |
| |FCP RELEASE | |
| |PME='dst udp 130.149.17.15:77 |
| | src udp 0.0.0.0:0 | |
| | if=in' | |
| | <---------------------- | |
J. Kuthan, J. Rosenberg [Page 20]
^LInternet Draft Middlebox Framework and Requirements November 2000
| | FCP OK | |
| | ----------------------> | |
| |FCP RELEASE | |
| |PME='dst udp 10.1.1.1:55 | |
| | src udp 0.0.0.0:0 | |
| | if=out' | |
| | <---------------------- | |
| | FCP OK | |
| | ----------------------> | |
| |FCP release_bind | |
| | udp 10.1.1.1:55 | |
| | <---------------------- | |
| | OK | |
| | -------------------------------------------> |
| | 200 OK | |
Figure 4: Protocol Flow for "SIP Session over NAT"
B Bibliography
1 S. Bradner: " The Internet Standards Process -- Revision 3", RFC
1602, IETF, October 1996.
2 B. Carpenter: "Achitectural Principle of the Internet", RFC 1958,
IETF, June 1996.
3 B. Carpenter: "Internet Transparency", RFC 2775, IETF, February
2000.
4 P. Srisuresh: " Framework for interfacing with Network Address
Translator", IETF, Internet Draft, July 2000. Work in progress.
5 M. Handley, H. Schulzrinne, E. Schooler, and J. Rosenberg: "SIP:
Session Initiation Protocol", RFC 2543, IETF, March 1999.
6 ITUi-T Recommendation H.323. "Packet-based Multimedia
Communications Systems," 1998.
7 H. Schulzrinne, A. Rao, R. Lanphier: "Real Time Streaming
Protocol", RFC 2326, IETF, April 1998.
8 Postel, J. and J. Reynolds, "File Transfer Protocol (FTP)", RFC
959, IETF. October 1985.
9 Schulzrinne, Casner, Frederick, Jacobson: "RTP: A Transport
Protocol for Real-Time Applications", Internet Draft, Internet
Engineering Task Force, March 2000, Work in progress.
10 P. Srisuresh and M. Holdrege: "IP network address translator
(NAT) terminology and considerations", RFC 2663, IETF, August
1999.
11 G. Tsirtsis, P. Srisuresh: "Network Address Translation -
Protocol Translation (NAT-PT)", RFC 2766, IETF, February 2000.
12 A. Feldmann, S. Muthukrishnann: "Tradeoffs for Packet
Classification", In Proc. IEEEi INFOCOM 2000, 2000.
13 V. Srinivasan, S. Suri, G. Varghese: "Packet Classification Using
Tuple Space Search", In Proc. ACM SIGCOMM '99, 1999.
J. Kuthan, J. Rosenberg [Page 21]
^LInternet Draft Middlebox Framework and Requirements November 2000
14 P. Gupta, N. McKeown: "Packet Classification on Multiple Fields",
In Proc. ACM SIGCOMM '99, 1999.
15 J. Rosenberg, D. Drew, H. Schulzrinne: "Getting SIP through
Firewalls and NATs", Internet Draft, Internet Engineering Task
Force, Feb. 2000. Work in progress.
16 M. Shore: "H.323 and Firewalls: Problem Statement and Solution
Framework", Internet Engineering Task Force, Feb. 2000. Work in
progress.
17 P. Ferguson, D. Senie: "Network Ingress Filtering: Defeating
Denial of Service Attacks which Employ IP Source Address
Spoofing", RFC 2827, IETF, May 2000.
C Glossary of Abbreviations
ACL Access Control List
ALGi Application Level Gateway
DMZ Demilitarized Zone
FCP Flow Processing Control Protocol
FTP File Transfer Protocol
IP Internet Protocol
HTTP Hypertext Transfer Protocol
MGCP Media Gateway Control Protocol
NAPT Network Address Port Translation
NAT Network Address Translation
NAT-PT Network Address Translation - Protocol Translation
RTP Transport Protocol for Real-time Applications
RTSP Real Time Streaming Protocol
RTT Round Trip Time
SDP Session Description Protocol
SIP Session Initiation Protocol
TCP Transmission Control Protocol
TOS Type of Service
UDP User Datagram Protocol
D Authors' Addresses
Jiri Kuthan
GMD Fokus
Kaiserin-Augusta-Allee 31
D-10589 Berlin, Germany
E-mail: kuthan@fokus.gmd.de
Jonathan Rosenberg
dynamicsoft
200 Executive Drive
Suite 120
West Orange, NJ 07052
email: jdrosen@dynamicsoft.com
J. Kuthan, J. Rosenberg [Page 22]
^LInternet Draft Middlebox Framework and Requirements November 2000
E Full Copyright Statement
Copyright (c) The Internet Society (2000). All Rights Reserved.
This document and translations of it may be copied and furnished
to others, and derivative works that comment on or otherwise explain
it or assist in its implementation may be prepared, copied,
published and distributed, in whole or in part, without restriction
of any kind, provided that the above copyright notice and this
paragraph are included on all such copies and derivative works.
However, this document itself may not be modified in any way, such
as by removing the copyright notice or references to the Internet
Society or other Internet organizations, except as needed for the
purpose of developing Internet standards in which case the
procedures for copyrights defined in the Internet Standards process
must be followed, or as required to translate it into languages
other than English.
The limited permissions granted above are perpetual and will not
be revoked by the Internet Society or its successors or assigns.
This document and the information contained herein is provided on
an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET
ENGINEERING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE
INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
J. Kuthan, J. Rosenberg [Page 23]
^L